# Gemini Tracking: How to Measure Brand Citations in Gemini

> Gemini tracking guide: build a prompt set, repeat runs, log cited URLs and read referrals now that Gemini dropped UTM tags. Steps and a free tool.

- Author: Sheetal Shakshi, Founder's Office, Nurdd
- Published: 2026-11-01
- Publisher: Noma by Nurdd (https://nomahq.app)
- Original page: https://nomahq.app/blog/gemini-tracking-brand-citations

Gemini tracking means running a fixed set of buyer questions through Gemini on a schedule, repeating each one, and recording whether your brand is named and which of your URLs are cited. Do not rely on UTM tags or analytics referrals: Gemini appears to have stopped adding *utm_source=gemini* to cited links in late September 2026.

## Key takeaways

- Track Gemini from the answer side: a fixed prompt set, repeat runs, the brand named or not, and every cited URL logged with the date.
- One check proves nothing. [Ahrefs, October 2026](https://ahrefs.com/blog/how-stable-are-ai-citations/) found about 80% of Gemini citations gone by the next day, so report a rate over many runs, not a screenshot.
- Gemini UTM tags came and went within weeks, so treat referral data as a partial, lower-bound signal, not your citation count.
- Pair answer-side tracking with server logs: the Gemini iOS app identifies itself in the user agent, while GA4 files many app visits under Direct.
- Start the prompt list with the [free AI Prompt Finder](https://nomahq.app/tools/ai-prompt-finder), then freeze it so every number stays comparable.

## How do I track my brand's citations in Gemini?

You track them in five steps: write a prompt set, freeze it, run each prompt several times in fresh sessions, log the answer text and every cited URL, then report rates with the denominator shown. This is the method behind any [Gemini visibility tracker](https://nomahq.app/features/gemini-visibility), whether you run it by hand or with software. For the wider metric set, see our guide to [tracking your brand in Gemini](https://nomahq.app/blog/gemini-visibility-tracker).

### Step 1: Build a prompt set that reflects real buying questions

Aim for 20 to 50 prompts, grouped by intent so a strong result in one group cannot hide a weak one in another. Ahrefs makes the same point: being cited for your own pricing is not the same as appearing for a "best X" question, so track factual brand questions, reviews and recommendations separately.

| Prompt group | Example pattern | What it tells you |
| --- | --- | --- |
| Category discovery | "What are the best [category] tools for [use case]?" | Whether Gemini puts you on the shortlist |
| Comparison | "[Your brand] vs [competitor] for [use case]" | How Gemini frames you against a named rival |
| Brand facts | "How much does [your brand] cost?" | Whether the facts Gemini states are right |
| Problem-led | "How do I fix [problem your product solves]?" | Whether your guides are cited before a buyer knows your name |

The [free AI Prompt Finder](https://nomahq.app/tools/ai-prompt-finder) shows the questions buyers ask ChatGPT, Gemini and Google AI about your category, tagged by intent, which gives you a starting list instead of a blank page. Our chapter on [choosing prompts that measure something](https://nomahq.app/aeo-guide/choosing-prompts) covers the five prompt kinds in more depth.

### Step 2: Freeze the set and write down the conditions

Version the list. If you reword a prompt, it becomes a new prompt with a new baseline. Record the date, the Gemini surface (web, iOS or Android app), the signed-in state, the country and the model name Gemini shows. A change in any of these can move results, and you will not know which one did it if you did not write it down. The [AI Visibility Measurement Standard](https://nomahq.app/resources/ai-visibility-measurement) sets out the same discipline: fixed, versioned prompts, each figure with its numerator and denominator.

### Step 3: Repeat every prompt, in fresh sessions

Run each prompt at least five times, each in a new chat with no history. Gemini's answer for a prompt is not one fixed result, so a single run tells you what happened once. Five runs across five days tells you how often it happens.

### Step 4: Capture the cited URLs, not just the brand name

A mention and a citation are different outcomes. Log them in separate columns. Our note on [citation versus brand mention](https://nomahq.app/blog/citation-vs-brand-mention) explains why: a brand can be named with no link, and a page can be cited while a competitor gets the recommendation.

For each run, record:

1. Prompt ID and run number
2. Whether your brand is named (yes or no) and its position among the brands listed
3. Every cited URL, copied in full, with the domain split out
4. Whether any cited URL is yours, and which page
5. The wording used about your brand, quoted
6. Competitors named

### Step 5: Report rates with the denominator

Report "named in 14 of 50 runs (28%)", not "28% visibility". Calculate three numbers per prompt group: mention rate (runs naming you divided by runs), citation rate (runs citing one of your URLs divided by runs) and top cited domains (a count by domain across all runs). The third often shows the real work: the pages Gemini leans on may be review sites or forums, not yours.

## Why does one Gemini check mislead you?

One check misleads because Gemini citations churn. In a 20-day study, Ahrefs followed about 37,500 citations across five AI tools, including Gemini (1,276 questions and 1,994 websites cited on day one).

**Finding:** at least half of AI citations disappeared by the next day across the five tools, about 80% on Gemini and Copilot, according to [Ahrefs, October 2026](https://ahrefs.com/blog/how-stable-are-ai-citations/).

Part of that Gemini figure is an artefact: the next answer often cited nobody. Excluding those answers, Ahrefs reports all five tools looked similar, with 50 to 60% of citations gone by the next answer that included citations. It also found that about 75% of the citations that dropped came back within 20 days, and that a citation holding for eight or more days almost always returned (96 to 99%).

Three rules follow from that. Count how many of the last seven days you were cited, as Ahrefs suggests. Judge a drop against what came before it. And do not credit a page edit for a single comeback. Our write-up of [AI citation stability](https://nomahq.app/blog/ai-citation-stability) turns these findings into a sampling plan.

| Reading | Weak method | Stronger method |
| --- | --- | --- |
| Was I cited? | One run, one screenshot | Cited in N of last 7 daily runs |
| Did a fix work? | Cited the day after the edit | Sustained rise in rate over several weeks |
| Is a drop real? | Absent today | Absent after a citation that had held 8+ days |
| Which pages matter? | Pages you expect | Cited URLs counted by frequency across runs |

## Did Gemini stop adding UTM parameters to links?

It appears so, though Google has not documented either the addition or the removal. Search Engine Journal reported on 1 October 2026 that Gemini had quietly started adding UTM parameters to outgoing links. Search Engine Roundtable later reported that Google had stopped around 28 September, with *utm_source=gemini* having grown for about three weeks and then vanished overnight, based on a graph from analyst Natzir.

Search Engine Roundtable's report that Gemini stopped attaching UTM parameters to cited URLs. Source: Search Engine Roundtable.

The two reports disagree on the timing, which is itself the lesson: undocumented behaviour can switch on and off without notice. Search Engine Journal noted that Google's implementation was undocumented, so it was unknown when Gemini added the tag. John Mueller of Google replied on Reddit when the tags were spotted:

> "It would be great to have these retained (in addition to the UTM-tagging)." [John Mueller, Google, quoted by Search Engine Journal, October 2026](https://www.searchenginejournal.com/google-gemini-adds-utm-parameters-for-referral-attribution/591754/)

Mueller was referring to the referrer header. According to Search Engine Roundtable, Gemini on desktop web still passes a *gemini.google.com* referrer, and visits with that referrer stayed flat while the UTM tags came and went. The same report says the Gemini iOS app opens links in its own WebView with no referrer, so in GA4 those visits land in Direct.

### What this changes for your measurement

| Surface | Referrer | UTM (since late September) | Where it lands in GA4 |
| --- | --- | --- | --- |
| Gemini on desktop web | Passed (gemini.google.com) | No longer seen, per Roundtable | Referral, AI assistants channel |
| Gemini iOS app | Not passed | No longer seen | Direct; user agent visible in server logs |
| Android assistant invocations | Reported stripped by one Reddit user, quoted by Search Engine Journal | No longer seen | Likely Direct (unverified) |

Do not build a Gemini citation KPI on UTM-tagged sessions. If you filtered a GA4 report on *utm_source=gemini* in September, expect a cliff in late September that says nothing about your visibility. Mark the date in your annotations.

## How do you measure Gemini traffic without UTM tags?

Use three sources side by side, because none is complete. The gap is large: in a server-log study, Wheelhouse DMG found 56 Gemini iOS visits in logs where GA4 recorded only 5 referrals, about 9% of the actual picture. Its authors call two accounts and one platform "early findings", so treat the ratio as a floor for one site type, not a rule.

1. **GA4 referrals.** Filter session source for *gemini.google.com*. This captures desktop web clicks only.
2. **Server logs.** According to [Wheelhouse DMG, March 2026](https://www.wheelhousedmg.com/insights/articles/ai-traffic-is-already-in-your-analytics/), the Gemini iOS app has identified itself with *GeminiiOS* and *GoogleWv* in the user agent since around 17 February 2026. Pull those requests and compare them with GA4 for the same dates.
3. **Direct baselining.** Chart monthly Direct sessions to deep content URLs that nobody types. A rise with no change in paid, email or other drivers is a lead worth investigating, not proof.

Run a manual test too. Open Gemini on your phone, ask a question that should surface your page, tap the link and watch your server log for the referrer. It takes ten minutes and tells you which case your site is in.

### Do not use Search Console as a Gemini report

Google's Search Console generative AI performance report, rolled out to all sites worldwide on 31 August 2026, covers AI Overviews and AI Mode. According to [Google Search Console Help, August 2026](https://support.google.com/webmasters/answer/16984139?hl=en), those are the capabilities included. The Gemini app is not listed, so the report cannot confirm Gemini citations. Use it for Google Search surfaces and keep Gemini on the answer-side method above. The [free Google Rank vs AI Mode Checker](https://nomahq.app/tools/google-rank-vs-ai-mode) lets you compare your ranking with what Google AI cites for one query.

## What does a worked example look like?

The numbers below are illustrative, not data from any brand. A team tracks 20 category prompts on Gemini, five runs each, so 100 answers.

| Metric | Calculation | Result |
| --- | --- | --- |
| Mention rate | Brand named in 31 of 100 answers | 31% |
| Citation rate | An owned URL cited in 12 of 100 answers | 12% |
| Named, not cited | 31 minus overlap of 9 answers | 22 answers to investigate |
| Top cited domain | Third-party review site in 40 answers | The page to earn a mention on |

Read the gap between 31% and 12% as a work list: in 22 answers Gemini names the brand but sends readers elsewhere, so the team checks which pages it cites instead and whether its own page answers that question plainly. Next month the team compares rates across the same 100 runs. A move from 31% to 34% is inside the noise of a sample this size, so it is not reported as a win.

## What are the common mistakes in Gemini tracking?

- **Changing prompts mid-stream.** Reworded prompts break comparisons. Add new prompts, never edit old ones.
- **Averaging across engines.** Gemini answers differently from ChatGPT or Perplexity. Report each engine separately; see [why engines disagree about your brand](https://nomahq.app/blog/why-engines-disagree).
- **Counting only branded prompts.** Prompts that include your name flatter your results. Most buyers start unbranded.
- **Treating UTM sessions as the citation count.** Sessions measure clicks, a fraction of influence, and the tag has been unstable.
- **Logging the brand but not the URLs.** Without cited URLs you cannot see which pages carry the answer.
- **Judging a fix after one good day.** Citations disappear and return as a matter of course.

## How does Noma fit into Gemini tracking?

Noma by Nurdd is an AEO (answer engine optimisation) and AI visibility platform. It asks 13 AI engines, including ChatGPT, Gemini, Perplexity and Claude, the questions your buyers ask every day from fresh sessions, and shows where your brand is named, which competitors appear instead and which pages the answers cite. Its agent analytics also report which AI crawlers fetch each page and which assistants send visitors back.

See it on your own brand: [book a demo](https://nomahq.app/demo) or [start the 3-day free trial](https://platform.nomahq.app/signup). A card is taken at signup so billing can start on day 3; cancel before then and nothing is charged. Read how [prompt tracking](https://nomahq.app/capabilities/prompt-tracking) and [agent analytics](https://nomahq.app/capabilities/agent-analytics) work, or see [pricing](https://nomahq.app/pricing).

## A checklist you can run this week

1. Generate 20 to 50 prompts with the free AI Prompt Finder and tag each by intent.
2. Freeze the list in a spreadsheet with an ID, version and date.
3. Run each prompt five times in fresh Gemini chats and log mentions and cited URLs.
4. Calculate mention rate, citation rate and top cited domains per group.
5. Pull *GeminiiOS* and *GoogleWv* requests from server logs and compare with GA4.
6. Annotate late September 2026 in analytics as the point Gemini UTM tags stopped appearing.
7. Repeat weekly and report a seven-day count, not a single result.

For attribution beyond the answer layer, our [AI search attribution guide](https://nomahq.app/blog/ai-search-attribution-guide) sets out how to connect visits to revenue.

## How we researched this

We reviewed Google's own Search Console documentation, trade press reports on the Gemini UTM change (Search Engine Journal and Search Engine Roundtable), a published 20-day citation-stability study from Ahrefs, a server-log analysis from Wheelhouse DMG, and the Gemini tracking guides currently ranking for this topic. Facts, dates and figures were checked on 2026-10-31. We have not run our own Gemini experiment for this article, and the worked example is illustrative. The Gemini UTM behaviour is undocumented by Google and may change again.

## What to do next

Build the prompt set today, run the first five-run baseline this week and keep the list frozen for at least a month. If you would rather not run hundreds of answers by hand, [book a demo](https://nomahq.app/demo) or [start the free trial](https://platform.nomahq.app/signup) and run the same prompts across 13 AI engines.